Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-supervised Learning with a Teacher-student Network for Generalized Attribute Prediction
Minchul Shin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add focal term (gamma=1) to teacher loss (ablation) | Ours(T) · DeepFashion Category | R@3: 88.45 -> 89.32 | Improved |
| tune gamma per attribute type for teacher (ablation) | Ours(T) · DeepFashion Category | R@3: 89.32 -> 90.01 | Improved |
| use label embedding instead of binary cross-entropy (ablation) | Ours teacher · FiccY | F1@1 avg: +0.62% | Improved |
| distill student from teachers | Ours(S) vs Ours(T) · iMatFashion Gender | F1@1: 88.50 -> 90.46 | Improved |
| distill student from teachers | Ours(S) vs Ours(T) · iMatFashion Category | F1@1: 42.50 -> 42.43 | No clear effect |
| distill student from teachers | Ours(S) vs Ours(T) · FiccY Texture | F1@1: 57.74 -> 59.19 | Improved |
| supervise student with four teachers instead of one (ablation) | Ours(S) · FiccY Texture | R@3: 93.57 -> 94.22 | Improved |
| supervise student with four teachers instead of one (ablation) | Ours(S) · FiccY Category | R@3: 97.30 -> 97.30 | No clear effect |
| add unlabeled target-domain images (DARN) to student training | Ours(S) · DARN cross-domain, DeepFashion train | F1@1 All: 65.15 -> 65.49 | Improved |
| add unlabeled target-domain images (DARN) to student training | Ours(S) · DARN cross-domain, iMatFashion train | F1@1 All: 73.10 -> 73.22 | Improved |
| mix 1M corrupted images into student training | Ours(S) · DeepFashion corrupted | R@3: 42.96 -> 43.96 | Improved |
| mix corrupted images into student training | Ours(S) · DeepFashion clean | R@3: 57.47 -> 57.80 | Improved |
Closest papers by meaning
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.88)
- Learning Attributes Equals Multi-Source Domain Generalization (CVPR 2016 · similarity 0.88)
- Efficient Relative Attribute Learning using Graph Neural Networks (ECCV 2018 · similarity 0.87)
- Semi-Supervised and Unsupervised Deep Visual Learning: A Survey (TPAMI 2022 · similarity 0.87)
- Label-Embedding for Image Classification (TPAMI 2016 · similarity 0.87)
- Unbiased Teacher for Semi-Supervised Object Detection (ICLR 2021 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
